FEASIBILITY OF ENVISAT DATA FOR THE ESTIMATION OF SNOW PACK CHARACTERISTICS AND AREAL FRACTION OF SNOW IN BOREAL FORESTS Jouni Pulliainen (1) , Sari Metsämäki (2) , Kari Luojus (1) , Martti Hallikainen (1) , Saku Anttila (2) , Juha-Petri Kärnä (1) , Markus Huttunen (2) , Sirpa Rasmus (3) , Jarkko Koskinen (4) , Tiia Grönholm (3) (1) Helsinki University of Technology, Laboratory of Space Technology, P.O. Box 3000, FIN-02015 HUT, Finland, Email: [email protected](2) Finnish Environment Institute (SYKE) (3) Department of Physical Sciences, University of Helsinki (4) Finnish Meteorological Institute ABSTRACT Estimation of snow moisture (total liquid water content) and the fraction of snow covered area (SCA) is investigated by applying multi-year ERS-2 SAR and Envisat ASAR data sets. Additionally, the feasibility of Envisat MERIS for the operative monitoring of SCA in boreal forests is discussed and analysed. An inversion approach for the moisture retrieval from SAR data is introduced. The results concerning space-borne radars suggest that C-band SAR is operationally feasible for both SCA and snow moisture monitoring, even though the accuracy of SAR-based SCA estimates is considerably poorer than that of optical image-based estimates. 1. INTRODUCTION Hydrological processes in boreal forest zone are highly affected by the seasonal snow cover. Thus, hydrological models operationally used for run-off and river discharge forecasting employ spatially distributed information on physical snow pack characteristics, and on the extent of snow. In Finland, the most important period is the spring melt season and the snow parameters essential for forecasts include the fraction of snow-covered area (SCA) and snow liquid water content (snow wetness). This information is required in a spatial scale from a few to several kilometres corresponding to sizes of sub-basins. The Finnish Watershed Simulation and Forecasting System (WSFS) applies snow information interpolated from weather stations and snow gauging network [1]. However, the accuracy of this interpolated information is relatively poor. Moreover, measurements on some important model parameters, such as snow liquid water content, are not carried out operationally. Space-borne observations can be used to overcome these problems [2,3,4]. The operational system of the Finnish Environment Institute (SYKE) already provides SCA- estimates derived from optical NOAA/AVHRR images during the spring melt period, and moreover, assimilates the obtained SCA estimates to WSFS system. However, these data are only available under non-cloudy conditions. Space-borne SAR provides information that can be used for the mapping of SCA regardless of cloud cover. SAR measurements can be also used to retrieve information on snow wetness. These aspects are investigated here. Additionally, the feasibility of Envisat MERIS full resolution data for snow monitoring, following the same procedure as with the AVHRR, is analysed. 2. MATERIAL AND METHODS 2.1 Test sites and data SCA retrieval, snow moisture mapping and the behaviour of microwave backscatter in boreal forest are investigated by applying extensive time-series of C- band ERS-2 SAR images from years 1997, 1998, 2000, 2001 and 2002, and additionally, Envisat ASAR images from the year 2003. NOAA/AVHRR band 1 (580-680 nm) data and Envisat/MERIS band 2 (442.5 nm) and band 6 (620 nm) data from melting period 2004 are applied to investigate the performance of MERIS in SCA estimation. AVHRR data are rectifies into the resolution of 1x1 km 2 , while full resolution MERIS data are rectified into the spatial resolution of 500 m. The main study area for SAR applications includes the eastern-most part of the River Kemijoki basin, while for optical investigations, the whole of Northern Finland is approached. The test sites around weather stations, selected for SAR analyses, are depicted in Fig. 1. Generally, the region represents a relatively flat relief, even though some mountainous areas (fjelds) are located in the region. The study area is almost totally covered by sparse conifer-dominated forests. Available reference data include (a) daily snow depth/conditions observations at weather stations, and (b) daily predictions on values of SCA, snow liquid water content and snow water equivalent (SWE) determined by WSFS. Additionally, detailed simulations on snow pack structure and its physical characteristics are obtained for a single test site (Sodankylä observatory, _____________________________________________________ Proc. of the 2004 Envisat & ERS Symposium, Salzburg, Austria 6-10 September 2004 (ESA SP-572, April 2005)
Transcript
FEASIBILITY OF ENVISAT DATA FOR THE ESTIMATION OF SNOW PACK CHARACTERISTICS AND AREAL FRACTION OF SNOW IN BOREAL FORESTS
Jouni Pulliainen(1), Sari Metsämäki(2), Kari Luojus(1), Martti Hallikainen(1), Saku Anttila(2),
Juha-Petri Kärnä(1), Markus Huttunen(2), Sirpa Rasmus(3), Jarkko Koskinen(4), Tiia Grönholm(3)
(1) Helsinki University of Technology, Laboratory of Space Technology, P.O. Box 3000, FIN-02015 HUT, Finland, Email: [email protected]
(2) Finnish Environment Institute (SYKE) (3) Department of Physical Sciences, University of Helsinki
(4) Finnish Meteorological Institute
ABSTRACT Estimation of snow moisture (total liquid water content) and the fraction of snow covered area (SCA) is investigated by applying multi-year ERS-2 SAR and Envisat ASAR data sets. Additionally, the feasibility of Envisat MERIS for the operative monitoring of SCA in boreal forests is discussed and analysed. An inversion approach for the moisture retrieval from SAR data is introduced. The results concerning space-borne radars suggest that C-band SAR is operationally feasible for both SCA and snow moisture monitoring, even though the accuracy of SAR-based SCA estimates is considerably poorer than that of optical image-based estimates. 1. INTRODUCTION Hydrological processes in boreal forest zone are highly affected by the seasonal snow cover. Thus, hydrological models operationally used for run-off and river discharge forecasting employ spatially distributed information on physical snow pack characteristics, and on the extent of snow. In Finland, the most important period is the spring melt season and the snow parameters essential for forecasts include the fraction of snow-covered area (SCA) and snow liquid water content (snow wetness). This information is required in a spatial scale from a few to several kilometres corresponding to sizes of sub-basins. The Finnish Watershed Simulation and Forecasting System (WSFS) applies snow information interpolated from weather stations and snow gauging network [1]. However, the accuracy of this interpolated information is relatively poor. Moreover, measurements on some important model parameters, such as snow liquid water content, are not carried out operationally. Space-borne observations can be used to overcome these problems [2,3,4]. The operational system of the Finnish Environment Institute (SYKE) already provides SCA-estimates derived from optical NOAA/AVHRR images during the spring melt period, and moreover, assimilates the obtained SCA estimates to WSFS
system. However, these data are only available under non-cloudy conditions. Space-borne SAR provides information that can be used for the mapping of SCA regardless of cloud cover. SAR measurements can be also used to retrieve information on snow wetness. These aspects are investigated here. Additionally, the feasibility of Envisat MERIS full resolution data for snow monitoring, following the same procedure as with the AVHRR, is analysed. 2. MATERIAL AND METHODS 2.1 Test sites and data SCA retrieval, snow moisture mapping and the behaviour of microwave backscatter in boreal forest are investigated by applying extensive time-series of C-band ERS-2 SAR images from years 1997, 1998, 2000, 2001 and 2002, and additionally, Envisat ASAR images from the year 2003. NOAA/AVHRR band 1 (580-680 nm) data and Envisat/MERIS band 2 (442.5 nm) and band 6 (620 nm) data from melting period 2004 are applied to investigate the performance of MERIS in SCA estimation. AVHRR data are rectifies into the resolution of 1x1 km2, while full resolution MERIS data are rectified into the spatial resolution of 500 m. The main study area for SAR applications includes the eastern-most part of the River Kemijoki basin, while for optical investigations, the whole of Northern Finland is approached. The test sites around weather stations, selected for SAR analyses, are depicted in Fig. 1. Generally, the region represents a relatively flat relief, even though some mountainous areas (fjelds) are located in the region. The study area is almost totally covered by sparse conifer-dominated forests. Available reference data include (a) daily snow depth/conditions observations at weather stations, and (b) daily predictions on values of SCA, snow liquid water content and snow water equivalent (SWE) determined by WSFS. Additionally, detailed simulations on snow pack structure and its physical characteristics are obtained for a single test site (Sodankylä observatory,
_____________________________________________________Proc. of the 2004 Envisat & ERS Symposium, Salzburg, Austria 6-10 September 2004 (ESA SP-572, April 2005)
2.2 Snow moisture retrieval from SAR data site 3 in Fig. 1) by applying the SNOWPACK model [5] developed by SLF (Swiss Federal Institute for Snow and Avalanche Research), see Fig. 2. In situ observation campaigns of snow pack characteristics were carried out for selected periods concurrently with satellite data acquisition. Thus, simulated values of snow grain size, stratification of snow pack, snow density, snow depth and snow moisture are validated against the measurement data. In situ snow moisture measurements were performed using a dielectric measurement device (Snow Fork).
An inversion method for deriving snow liquid water content information from C-band radar images is introduced in this investigation. The performance of this approach is assessed by comparing the SAR-derived estimates with hydrological model predictions. The investigated snow wetness characteristic is the total amount of liquid water in the snow pack (in mm). The vertical snow moisture profile is typically very heterogeneous as liquid water is concentrated to a certain distinct layer (see Fig. 2). Due to this, SAR estimated snow wetness value should not be compared with the average moisture of the snow pack, but instead with the total amount of liquid water.
The applied methods for estimating both snow wetness and SCA are based on the semi-empirical modeling of C-band backscattering coefficient as a function of forest cover (biomass level), soil properties and snow pack characteristics [6,7,8].
The estimation of liquid water content of snow is restricted to cases representing 100% snow cover. The snow backscattering model is a discrete scattering approach based semi-empirical model described in [8]. In this model, the snow-air surface backscattering contribution, essential under the wet snow conditions, is treated with the IEM model [9]. The soil backscattering model used for modeling the backscatter from snow-free ground (and the snow ground interface in the snow model) is the empirical model from [10]. Altogether, the snow backscattering model is simple enough to be applied in iterative inversion by using a constrained non-linear minimization procedure. In the case of snow liquid water content estimation from ERS-2 SAR and Envisat ASAR data we can write the estimation algorithm as
Fig. 1. The main study area and the locations of the 14
km by 15 km-sized test regions. The land-use classes of the region are coded with different colors: water areas
(blue), open areas (brown), mires and bogs (bluish green), dense forests (dark green) and sparse forests
(light green).
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 5.50
20
40
60
Hei
ght (
cm)
Snow Fork Observations (6 April 2003)
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 5.50
20
40
60
Hei
ght (
cm)
Model Prediction (27 April 2003)
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 5.50
20
40
60
Snow Moisture (%)
Hei
ght (
cm)
Model Prediction (24 April 2002)
( )ggsnowvsnow
snowsurfm
sdlsmf
snowv
ερσ
σσ
,,,,,,
where
min
0,
,
=°
°−°
(1)
where snow°σ is the modeled backscattering coefficient of snow covered ground; and the parameters of snow model include, in addition to snow wetness mv,snow, snow surface rms height s, horizontal snow surface correlation length l, effective snow grain size d0, snow density ρ, ground surface rms height s, soil dielectric constant εg (the used average values are 6 mm, 5 cm, 2.8 mm, 0.3 g/cm3, 1.2 cm, 6-1*j, respectively). The ground surface backscattering contribution surf°σ (scalar variable) is derived directly from SAR data. It is also estimated for forested areas assuming that the whole ground is covered by snow. The method used for
Fig. 2. Vertical snow moisture profiles for the
Sodankylä site (Sodankylä observatory) corresponding to three SAR image aqcuisition dates. Above: Snow
profiles observed with dielectric measurement system. Middle and below: SNOWPACK model simulations. The profile obtained for 24 April 2002 corresponds to
2.4 SCA retrieval from optical data compensating the effects of forest cover is described in [7].
The SCA estimation from optical data is based on the semi-empirical model, where reflectance from a target area is expressed as a function SCA:
2.3 SCA retrieval from SAR data
The performance of SAR-based SCA estimation in forested areas is analyzed by comparing the obtained estimates with independent SCA reference data sources: (a) hydrological model predictions and (b) observations at weather stations.
[ ]groundsnow
forestobs
ρSCAρSCAt
ρtSCAρ
,,2
,2
,
)1(
)1()(
λλλ
λλλ
∗−+∗+
∗−= (2)
where ρλ,snow, ρ λ,ground and ρλ,forest are the generally
applicable reflectances for wet snow, snow-free ground and dense coniferous forest canopy at wavelength λ, respectively. ρλ,obs stands for observed reflectance from areal calculation unit with current snow cover fraction. tλ stands for effective transmissivity within the calculation unit. Through transmissivity, the forest coverage for each calculation unit is individually determined. The key idea is that the effective transmissivity is estimated using the Earth observation data, and even using the similar data as employed in the actual SCA estimation [11]. This kind of approach enables operational snow mapping in an extensive and heterogeneous area such as boreal zone.
SCA estimates, 28 May 1997.
0 km 10 km 20 km 30 km 40 km 50 km 60 km 70 km
70 km
60 km
50 km
40 km
30 km
20 km
10 km
0 km
SCA [%]
0
10
20
30
40
50
60
70
80
90
100
In the operational SCA monitoring, the drainage basins of Finland (average acreage of 60 km2) are used as calculation units. Basin mean reflectance serves as reflectance observation for the SCA model, which gives the average SCA for the basin. The transmissivities are calculated using AVHRR reflectances under full dry snow cover conditions. For MERIS SCA estimation in this study, there were not sufficiently data representing full dry snow cover conditions. Therefore, the transmissivitites applied for MERIS observations were derived from Terra/MODIS 500m-data, as their resolution is closer to MERIS data than to AVHRR data.
Fig. 3. An example on ERS-2 SAR intensity image
( °σ -values) together with SAR-based SCA estimates determined from it for 14 sub-drainage areas of the Lokka region (vicinity of test site 2, see Fig. 1). The
increase in the areal fraction of snow is also visible in the change of backscattering intensity: Dark colours
indicate low °σ -values, and thus, high levels of SCA. The SCA estimation from SAR data is based on the linear interpolation between reference images representing 100% wet snow cover and totally snow-free conditions [2]. However, in forested areas the backscattering and attenuation effects of forest cover are eliminated prior to the linear interpolation by applying a non-linear forest compensation method [7]. In this approach a non-linear forest backscattering model [6] is fitted to SAR data in order to estimate the level of backscatter originating from the forest floor (forest floor can be totally or partially snow covered of snow-free in the image under investigation). The algorithm requires that forest stem volume/biomass information with a high spatial resolution is available for the region under study, as the model fitting has to be carried out using mean σ°-values calculated for various stem volume classes. Fig. 3 shows an example of SCA retrieval for forests in the sub-drainage areas of the Lokka region.
As the SCA-model uses generally applicable average reflectances for wet snow, bare ground and forest canopy, choosing the best possible wavelength is crucial for the success of the SCA estimation. For wet snow, small wavelengths are suitable, as the effect of grain size increases towards larger wavelengths. Furthermore, as the appearance of seasonal green vegetation may disturb the estimation at the end of the melting season with patchy snow cover, it is important to use a band insensitive to vegetation. The MERIS band meeting these criteria is band 2 (442.5 nm). Hence, the value for ρλ,snow, was derived from MERIS band 2 data representing wet snow, and the values for ρ λ,ground and ρλ,forest were derived from band 2 data under melt-off conditions using pixels representing open bog and dense coniferous forests, respectively. Basin mean reflectance at band 2 represents the reflectance observation.
Fig. 4. depicts a time series of MERIS band 2 reflectances for an arbitrary basin with relatively dense forest coverage. AVHRR band 1 reflectances are presented for comparison, as well as the behaviour of MERIS band 6, which is the closest one to the AVHRR band 1. Fig. 4 indicates that with almost full snow cover, the MERIS band 2 reflectance shows the highest value, whereas it shows the lowest value for snow-free terrain. The reflectances also illustrate that the disturbing effect of emerging green vegetation is stronger for the band 1 of AVHRR than for the MERIS band 2. Thus, the dynamic response of MERIS band 2 to SCA is higher than that of AVHRR band 1, which indicates an improved feasibility for snow mapping
Fig. 4. Tim
applications.
e series of MERIS bands 2 and 6 and
. RESULTS AND DISCUSSION
.1 Radar applications
his section presents examples of results obtained in
he liquid water content of snow pack (snow wetness)
AVHRR band 1 for a basin with moderate forest coverage.
3 3 Tsnow liquid water content estimation using ERS-2 SAR and Envisat ASAR data. Additionally, examples on results achieved in SCA estimation are presented. Twas estimated for SAR images acquired for conditions with a full snow cover. In Figs. 5 and 6, the comparison of estimation results with hydrological model predictions indicates that the liquid water content of snow pack can be mapped in a regional scale of few kilometers both for open and forested areas even using a single channel system (for forested areas reference information on biomass is required). Figs. 5 and 6 show that C-band SAR-based snow liquid water content estimates coincide relatively well with the six-
year hydrological model predictions in case of all applied test areas, each sized 15 km by 15 km. The temporal variation of snow grain size is a major source of error in SAR-based snow liquid water content estimation. In case of measured and simulated data presented in Fig. 2, the SNOWPACK model predicted grain size diameter varies under wet snow conditions from 0.93 to 2.20 mm in 24 April 2002, and from 1.02 to 3.20 mm in 27 April 2004. The snow grain size variation observed under dry snow conditions in 6 April 2003 is from 0.5 to 5 mm (5 mm for depth hoar). Fig. 7 depicts an example of Envisat ASAR-derived backscattering signatures for the Lokka and Sodankylä sites, regions 2 and 3 in Fig. 1, respectively. In addition to the observed average backscattering coefficients of different stem volume classes, the fitting of semi-empirical forest backscattering model [5] to observation is shown. The total snow liquid water contents estimated from SAR data show a value of 3.3 mm for the Lokka site, and a value of 4.7 mm for the Sodankylä site. The values obtained from WSFS simulations are 1.8 mm and 2.7 mm, respectively. The concurrent measured snow moisture profile for the Sodankylä station is depicted in Fig. 2. This profile indicates a considerably lower level of liquid water content (0.6 mm) than the WSFS simulation.
0 20 40 60 80 100 120 140 160
0
1
2
3
4
5
6
Julian Day in Year 2003
Sno
w L
iqui
d W
ater
Con
tent
(m
m)
Envisat ASAR− and WSFS−Derived Liquid Water
Fig. 5. Total snow liquid water content of snow pack simulated by the WSFS system (solid line) and
corresponding estimates using Envisat ASAR data (circles). The results are depicted for the Savukoski
Fig. 6. Examples on WSFS hydrological forecasting and simulation system predicted total liquid water content of snow (solid line) and the corresponding ERS-2 SAR based estimates (asterisks). Typical results are shown for some of the test
sites for the years 1997, 1998, 2000, 2001 and 2002.
0 20 40 60 80 100 120 140 160 180 200−10.5
−10
−9.5
−9
−8.5
−8Lokka, 6 April 2003
σo (dB
)
VV pol.HH pol.
0 20 40 60 80 100 120 140 160 180 200−10.5
−10
−9.5
−9
−8.5
−8Sodankylä obs., 6 April 2003
Stem Volume (m3/ha)
σo (dB
)
VV pol.HH pol.
Fig. 7. Example on Envisat ASAR observations for Lokka and Sodankylä test sites for 6 April 2003 (cold winter conditions with snow depths of 53 and 64 cm, respectively). Circles and triangles show the observed average values for different stem volume classes, whereas solid lines show the modeled curve obtainde by fitting the semi-empirical model
into the observation data. Angle of incidence is 28.54° (beam center).
The performance of SCA retrieval was also investigated for the test sites depicted in Fig. 1. Again, the objective was to determine the average conditions for each 15 km by 15 km sized test site: The mean SCA of forested areas and that of the open areas, respectively. Figs. 8 and 9 show examples of SCA retrieval results for the spring of 1997. The overall performance of SCA retrival is summarized in Table 1. The results indicate the RMSE of SCA estimation when compared with the WSFS model predictions for melting periods of five years. The results presented in Table 1 and Figs. 8 and 9 are determined for the 15 km by 15 km-sized test sites surrounding weather stations (see Fig. 1).
0 50 1000
50
100
Forests (Comp.)
Est
. SC
A (
%)
0 50 1000
50
100
Forests (Lin.Interp.)
Est
. SC
A (
%)
0 50 1000
50
100
Open Areas
Est
. SC
A (
%)
Ref. SCA (%)
Fig. 9. Above: An example on SAR-based SCA retrieval results for a single test site, region 2 in Fig. 1.
The behaviour of WSFS-based reference values are also shown for forested (blue curve) and open areas (red curve). Below: Snow observations at the Lokka
weather station for the same period of time. The observed snow classes (snow codes) are: 7 = full wet
snow cover, 6 = wet snow with %100SCA%50 <≤ , 5 = wet snow with SCA < 50%, 4 = wet snow on forests,
0 = snow-free terrain. The results obtained in SCA mapping indicate that the SCA retrieval accuracy of C-band SAR is poorer than that of optical satellite instruments. The RMSE of SAR-based SCA estimates obtained in this investigation for regions around weather stations shows values in the order of 35%-units. The RMSE value obtained using the same multi-year data set for the 14 sub-drainage areas shown in Fig. 3 is around 20 %-units (depending on the method of reference image selection). The corresponding RMSE of NOAA AVHRR-based estimates is roughly 15%-units (as estimates a calculated for sub-drainage areas with a mean size of 36 km2). However, as optical images are not always available due to cloud cover, SAR data can be applied for those cases. The results also indicate that the semi-empirical backscattering model-based compensation [5,6] of forest canopy backscatter can reduce the disturbances caused by the forest cover, refer to Table 1.
Fig. 8. SAR-based SCA estimates for all 15 km by 15 km-sized test sites for the spring 1997. The SAR-based
estimates are comaperd with WSFS simulations calculated for the same regions. The RMS-errors of
estimates when compared with WSFS reference data are given in Table 1.
3.2 Applicability of optical instruments With the methodology described above in Section 2.4,
SCA estimates for 994 drainage basins were calculated for the MERIS image acquired on the 6th of May, 2004 (see Fig. 10). The estimation result for the same MERIS scene is shown in Fig. 11, where for comparison, SCA-estimates derived from operative AVHRR-based snow monitoring system are also presented. An RMSE of 15 % (SCA-percentages) for AVHRR-derived estimates in general was gained, when ground truth SCA from Finnish snow courses from years 2001-2003 were employed. Fig. 12 presents the comparison between MERIS-derived and AVHRR derived SCA-estimates indicating a small bias between the estimates. This bias is likely to occur because the effective transmissivities were derived using MODIS data instead of MERIS due to the lack of MERIS data.
Fig. 11. SCA-estimates for Finland calculated from MERIS (left) and from AVHRR (right) data. The areas marked with white colour are clouds or noise stripes in the original data (see northern Finland in the AVHRR-
derived map).
Fig. 12. Comparison between MERIS-derived and
AVHRR-derived SCA-estimates 4. CONCLUSIONS We have investigated the applicability of C-band SAR data for the estimation of snow liquid water content and SCA in boreal forest zone. The SCA retrieval in boreal areas requires a priori information on forest biomass (stem volume), and the estimation of snow liquid water content additionally requires information on snow pack thickness or snow water equivalent. The results indicate that C-band SAR is potential for operative use in case of both applications.
Figure 10. MERIS full resolution image on the 6th of May, 2004, bands 10, 15 and 9 (RGB). Especially on
open areas, snow and ice stand out in purple.
The results obtained for the optical data indicate that the MERIS instrument is able to provide a higher
performance for SCA mapping than the AVHRR, both in open and forested areas. The results suggest that, in addition to the effect of higher spatial resolution, the better spectral channel configuration of MERIS improves the SCA retrieval accuracy. 5. ACKNOWLEDGEMENTS This work is supported by the ESA Envisat AO-project "Assessment of the usability of ENVISAT MERIS, AATSR, ASAR data in monitoring of coastal waters, lakes and snow in Finland" (AO400), and by the EC EnviSnow project (Development of Generic Earth Observation based Snow Parameter Retrieval Algorithms, EVG1-CT-2001-00052). 6. REFERENCES 1. Vehviläinen , B., Snow Cover Models in Operational Watershed Forecasting, Publications of the Water and Environment Research Institute, No. 11, Helsinki, Finland, 1992. 2. Koskinen, J., Pulliainen, J. and Hallikainen, M., The use of ERS-1 SAR data in snow melt monitoring, IEEE Trans. Geosci. Remote Sensing, vol. 35, 601-610, 1997. 3. Nagler, T. and Rott, H., Retrieval of wet snow by means of multitemporal SAR data, IEEE Trans. Geosci. Remote Sensing, vol. 38, 754-765, 2000. 4. Metsämäki, S., Vepsäläinen, J., Pulliainen, J., and Sucksdorff, Y., An improved linear interpolation method for estimation of snow covered area, Remote Sensing of Environment, vol. 82, 64-78, 2002. 5. Bertelt, P. and Lehning, M., A physical SNOWPACK model for the Swiss avalanche warning: Part I. Numerical model, Cold Reg. Sci. Technol., vol. 35, 123-145, 2002.
6. Pulliainen, J., Kurvonen, L. and Hallikainen, M., Multi-temporal behavior of L- and C-band SAR observations of boreal forests, IEEE Trans. Geosci. Remote Sensing, vol. 37, 927-937, 1999. 7. Pulliainen, J., Koskinen, J. and Hallikainen, M., Compensation of forest canopy effects in the estimation of snow covered area from SAR data, Proc. IGARSS'01, Sydney, Australia, 9-13 July 2001. 8. Koskinen, J., Snow Monitoring Using Microwave Radars, Ph.D. Dissertation, Helsinki University of Technology, Laboratory of Space Technology, Report 44, Espoo, Finland, January 2001. 9. Fung, A., Microwave Scattering and Emission Models and Their Applications, Norwood: Artech House, 1994. 10. Oh, Y., Sarabandi, K. and Ulaby, F., An empirical model and an inversion technique for radar scattering from bare soil surfaces, IEEE Trans. Geo. Remote Sensing, vol. 30, 370-381, 1992. 11. Metsämäki, S., Huttunen, M. and Anttila, S., The operative remote sensing of snow covered area in a service of hydrological modelling in Finland, in Remote Sensing in Transition (proceedings of 23nd Symposium of EARSeL, Gent, Belgium, 2-5 June 2003). Millpress Rotterdam, 2004. ISBN 90 5966 007 2.
